Cufft error in file - cuda

Question

I am receiving the error:
Cufft error in file
I am using this file in order to load the FFT and pass them to another file.
//----function to check for errors-------------------------------------------------
#define gpuErrchk(ans) { gpuAssert((ans), __FILE__, __LINE__); }
inline void gpuAssert(cudaError_t code, char *file, int line, bool abort=true)
{
if (code != cudaSuccess)
{
fprintf(stderr,"\nGPUassert: %s %s %d\n", cudaGetErrorString(code), file, line);
if (abort) exit(code);
}
}
//function to check for cuFFT errors --------------------------------------------------
#define CUFFT_SAFE_CALL( call) do { \
cufftResult err = call; \
if (err != CUFFT_SUCCESS) { \
fprintf(stderr, "Cufft error in file '%s' in line %i : %s.\n", \
__FILE__, __LINE__, "error" ); \
exit(EXIT_FAILURE); \
} \
} while (0)
#define NX 128*128
#define NY 16
#define BATCH 16
#define NRANK 2
void FFT_transform(cufftDoubleComplex** B_in)
{
int n[NRANK] = {NX, NY};
//size of B
int Bsize=NX*NY*BATCH;
//allocate host memory
*B_in=(cufftDoubleComplex*)malloc(Bsize*sizeof(cufftDoubleComplex));
for (int i=0;i<NX*NY;i++){
for (int j=0;j<BATCH;j++){
(*B_in)[i*BATCH+j].x=(i*BATCH+j)*2;
(*B_in)[i*BATCH+j].y=(i*BATCH+j)*2+1;
}
}
//allocate device memory
cufftDoubleComplex* B_dev;
gpuErrchk(cudaMalloc((void**) &B_dev,Bsize* sizeof(cufftDoubleComplex)));
if (cudaGetLastError() != cudaSuccess){
fprintf(stderr, "Cuda error: Failed to allocate\n");
return;
}
// copy arrays from host to device
gpuErrchk(cudaMemcpy(B_dev, *B_in,Bsize* sizeof(cufftDoubleComplex), cudaMemcpyHostToDevice));
// Create a 2D FFT plan
cufftHandle plan;
CUFFT_SAFE_CALL(cufftPlan2d(&plan,NX,NY,CUFFT_Z2Z));
if (cufftPlanMany(&plan, NRANK, n,NULL, 1, 0,NULL, 1, 0,CUFFT_Z2Z,BATCH) != CUFFT_SUCCESS){
fprintf(stderr, "CUFFT Error: Unable to create plan\n");
return;
}
if (cufftSetCompatibilityMode(plan, CUFFT_COMPATIBILITY_NATIVE)!= CUFFT_SUCCESS){
fprintf(stderr, "CUFFT Error: Unable to set compatibility mode to native\n");
return;
}
// perform transform
CUFFT_SAFE_CALL(cufftExecZ2Z(plan,(cufftDoubleComplex *)(*B_in), (cufftDoubleComplex *)B_dev, CUFFT_FORWARD));
if (cufftExecZ2Z(plan,*B_in,B_dev,CUFFT_FORWARD) != CUFFT_SUCCESS){
fprintf(stderr, "CUFFT Error: Unable to execute plan\n");
return;
}
if (cudaThreadSynchronize() != cudaSuccess){
fprintf(stderr, "Cuda error: Failed to synchronize\n");
return;
}
// copy result from device to host
gpuErrchk(cudaMemcpy(*B_in, B_dev,Bsize*sizeof(cufftDoubleComplex), cudaMemcpyDeviceToHost));
//Destroy CUFFT context
CUFFT_SAFE_CALL(cufftDestroy(plan));
//clean up device memory
gpuErrchk(cudaFree(B_dev));
}
I am receiving the error at line:
CUFFT_SAFE_CALL(cufftExecZ2Z(plan,(cufftDoubleComplex *)(*B_in), (cufftDoubleComplex *)B_dev, CUFFT_FORWARD));

Answer 1

You are getting the error because B_in is a pointer to host memory and not to device memory, which is illegal. In CUFFT, inputs are always in device memory. You need to use cudaMemcpy to transfer the contents of B_in to B_dev before performing the transform, and then supply B_dev as both the input and output, which will result in an in place transform. This is clearly described in the CUFFT API documentation here.